The Physical Stack of AI · Sovereign AI compute
You can explain why the Gulf states and India built sovereign-AI strategies from scratch rather than out of legacy supercomputing — and why that gave them larger, faster announcements but also tougher political optics.
Where Europe wired AI compute into existing HPC centres, the Gulf and South Asia treated it as a greenfield national project, closer in shape to a new airport or port than a research grant. The biggest single deal is Stargate UAE — a 1 GW Abu Dhabi cluster, with the first 200 MW expected to go live in 2026, built by G42 and operated by OpenAI and Oracle alongside NVIDIA, Cisco, and SoftBank. Crucially, OpenAI frames it as the first international Stargate deployment and a government-coordinated OpenAI for Countries partnership, which is a deliberate signal about who counts as inside the allied compute tent.
Saudi Arabia's HUMAIN, anchored by a $10B PIF-Google joint venture for a Riyadh AI hub, is the parallel play. India is building from a different starting point: the IndiaAI Mission is using ₹10,372 crore (~$1.25B) in subsidies to bring GPU-hour costs down by roughly 42% under market, with the explicit aim of getting Indian startups access to Nvidia silicon at $1.50/hr-ish prices. Target: 100K GPUs in production by end of 2026, from a base of 18K in early 2025 and 34K in early 2026.
Chapter contains 4 lessons.